The HN article that this is about: https://news.ycombinator.com/item?id=23893817
Congratulations to the authors of these comments, who correctly guessed that it was written by GPT-3:
Congratulations to the authors of these comments, who correctly guessed that it was written by GPT-3:
However I would agree that it is an oblique version - that is, can a machine fool humans into thinking that they are human.
In which case I think it's probably safe to assume that GPT-3 has passed.
> The object of the game for the interrogator is to determine which of the other two is the man and which is the woman. He knows them by labels X and Y, and at the end of the game he says either "X is A and Y is B" or "X is B and Y is A." The interrogator is allowed to put questions to A and B
> What will happen when a machine takes the part of A in this game?" Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman?
Somewhat politically incorrect, assuming men and women should ever be distinguished, but much more revealing about how exactly people see themselves.
Exactly.
(If the commenter had written even a couple sentences on why they suspected GPT-3, I would feel differently.)
> This article is vapid. Zero substantive content, pure regurgitation
Now, I would still have downvoted such a comment. If I'm reading a thread, I clearly like at least the article's topic, and I want to read more on that topic, not dismissive comments telling me that it's stupid. But, at least the modified version focuses on the content, rather than the author.
I think we're likely to get to a place where we either consolidate to trusted sources of information, or accept machine generated content as valuable on its own, even if for novelty and just move on with our miserable lives.
I'm certain GPT-3 has been commenting on HN threads for a while now. In some cases, its presence has been disclosed (see, for example: https://news.ycombinator.com/item?id=23886503) In other cases, GPT-3's presence has not been disclosed; the machine has been pretending to be a human being, largely unnoticed. Consider only how easy it is for it to write short, punchy comments -- say, one to three sentences long.
By implication, there's a high probability that we -- you, me, and everyone else on HN -- have been upvoting and downvoting GPT-3 comments for a while without realizing it.
And the technology is only going to get better.
The example comments you linked (Where GPT-3's presence was disclosed) were believably human, particularly if you were skimming, but they were not good comments. If not for the note at the end about GPT-3, I'm pretty confident they would have been downvoted.
And if I'm wrong, and GPT-3 is actually capable of writing thoughtful and substantive comments... well, in the words of XKCD, "mission fucking accomplished."
To be fair though: it's quite clear that all the discussion in that post is commenting on the headline and not the content. Bland self-improvement "life hacks" are one of the metaphorical crack pipes of this site. We all have way too much to say on the subject of our own productivity.
So I think it's less likely that HN was fooled by GPT-3 but that GPT-3 was good enough at filling a plausible article out around a tempting headline.
GPT-3 fits snugly into the pattern of "AI does intelligent thing -> we decide/realize thing doesn't reflect meaningful intelligence". Maybe in this case the devaluing of blog "crack" is a positive thing.